Communications of the ACM
Google and Facebook Race to Solve the Ancient Game of Go With AI
Facebook and Google are using machine-learning techniques to try to develop software that can win at the ancient, complex game of Go.
Credit: Takashi Osato/Wired
Although software designed to play games such as checkers, chess, and some video games can now dominate even the best human players, the ancient game of Go has proven a much trickier challenge.
Played on a 19-by-19 board, this Asian version of chess features too many possible moves in any given turn for a current computer to quickly analyze all of the possibilities and their outcomes to find the optimal next move; this method, called a Monte Carlo tree search, has been the primary strategy for many game-playing systems based on artificial intelligence.
However, Google and Facebook are hoping to crack Go by taking a different approach. Both companies are turning to the machine-learning techniques they currently use to recognize and categorize pictures, translate text, and process voice commands. Their efforts are promising because they mimic the way professional Go players approach the game--a visual and instinctive approach that depends on identifying patterns, a process well-suited to deep-learning systems.
Facebook recently announced a new method blending deep learning with a Monte Carlo tree search, and Google has said it plans to announce its own breakthrough soon.
From Wired
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found